As you might imagine, mapping the universe on the largest of scales and tracing threads of the cosmic web is no mean feat.
To do so, you’d need to take observational evidence and combine it with theoretical models such as the Effective Field Theory of Large-Scale Structure (EFTofLSS). Only then would you be on the road to developing a statistical map of the universe’s 3D skeleton. This is a task way beyond human minds alone, requiring a vast amount of computing time — and that’s very valuable time, especially as catalogues of astronomical data continue to grow exponentially.
So, how can we cut down the time such analyses take without sacrificing precision? An international team of researchers thinks it has come up with a solution to this problem: an emulator dubbed “Effort.jl.” Effort.jl should, the team says, deliver the same precision of EFTofLSS while running on a laptop rather than a supercomputer and taking mere minutes.
“Imagine wanting to study the contents of a glass of water at the level of its microscopic components, the individual atoms, or even smaller: in theory, you can. But if we wanted to describe in detail what happens when the water moves, the explosive growth of the required calculations makes it practically impossible,” Marco Bonici, study team leader and a researcher at the University of Waterloo, said in a statement. “However, you can encode certain properties at the microscopic level and see their effect at the macroscopic level, namely, the movement of the fluid in the glass.
“This is what an effective field theory does, that is, a model like EFTofLSS, where the water in my example is the universe on very large scales and the microscopic components are small-scale physical processes.”
Theoretical models like EFTofLSS take astronomical data and make predictions that explain those datapoints. The problem is, however, that such surveys like that conducted by the Dark Energy Spectroscopic Instrument (DESI), which delivered its first results in April 2024, and the European Space Agency (ESA) spacecraft Euclid are composed of incredibly large data sets. That volume of data is impractical to fit theoretical models to deliver large-scale, accurate predictions.
“This is why we now turn to emulators like ours, which can drastically cut time and resources,” Bonici said.
Emulators like Effort.jl are built on neural networks trained using theoretical models, learning parameters, and predictions already made, meaning they can mimic the function of these models. That means while emulators can’t understand the physics they are dealing with, they can take a new input and output a prediction that conforms to what the traditional model itself would predict.
Effort.jl, in fact, takes this further by integrating the knowledge of how predictions will change if the parameters of a model are changed. This emulator can also account for how predictions will shift if parameters are merely tweaked by tiny amounts. This means Effort.jl can learn with fewer examples being fed to it than other emulators. This means it can run with less computing power.
The new research from Bonici and colleagues validates the accuracy of Effort.jl when tackling real astronomical data and simulated data, with predictions conforming closely to those produced by the EFTofLSS.
“In some cases, where with the model you have to trim part of the analysis to speed things up, with Effort.jl we were able to include those missing pieces as well,” Bonici added.
That makes Effort.jl a promising ally for next-generation cosmological efforts such as those conducted by DESI and Euclid in our ongoing effort to better understand the large-scale structure of our universe.
The team’s research was published on Tuesday (Sept. 16) in the Journal of Cosmology and Astroparticle Physics (JCAP).